From 1277fcda8e81fdbb19497679c7f2d403f7367864 Mon Sep 17 00:00:00 2001 From: Janek Date: Thu, 7 May 2026 12:59:19 +0200 Subject: [PATCH] fix configs --- configs/_cluster/entropy.yaml | 4 ++-- configs/_cluster/helios.yaml | 11 ++++++----- 2 files changed, 8 insertions(+), 7 deletions(-) diff --git a/configs/_cluster/entropy.yaml b/configs/_cluster/entropy.yaml index e8a3a2f0..105a5d99 100644 --- a/configs/_cluster/entropy.yaml +++ b/configs/_cluster/entropy.yaml @@ -12,7 +12,7 @@ infrastructure: script: - '${export_env_variables_placeholders:}' - - 'export PROJECT_HOME_PATH=/storage_nvme_4/nano/$USER' + - 'export PROJECT_HOME_PATH=/storage_nvme_2/nano/$USER' # hydra errors - 'export HYDRA_FULL_ERROR=1' @@ -34,4 +34,4 @@ cluster_switch: train_path_c4: "/storage_nvme_1/llm-random/datasets/c4/train" eval_path_c4: "/storage_nvme_1/llm-random/datasets/c4/validation" train_path_fineweb: "/storage_nvme_4/llm-random/datasets/fineweb/train" - eval_path_fineweb: "/storage_nvme_4/llm-random/datasets/fineweb/train" + eval_path_fineweb: "/storage_nvme_4/llm-random/datasets/fineweb/train" \ No newline at end of file diff --git a/configs/_cluster/helios.yaml b/configs/_cluster/helios.yaml index f130fe37..bac89976 100644 --- a/configs/_cluster/helios.yaml +++ b/configs/_cluster/helios.yaml @@ -33,17 +33,18 @@ infrastructure: - 'export UV_CACHE_DIR=$PROJECT_HOME_PATH/cache/uv' - 'export XDG_CACHE_HOME=$PROJECT_HOME_PATH/cache' - # Prepend env's libstdc++ so triton's libtriton.so resolves CXXABI_1.3.15 (GCC 14 ABI). - # Helios's system/module libstdc++ is too old and breaks torch.compile on aarch64. - - 'export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"' - # activate pixi - 'cd "$PIXI_HOME"' - 'eval "$(pixi shell-hook)"' - 'cd -' + # Prepend env's libstdc++ so triton's libtriton.so resolves CXXABI_1.3.15 (GCC 14 ABI). + # Helios's system/module libstdc++ is too old and breaks torch.compile on aarch64. + # Must come AFTER pixi shell-hook — that's what sets CONDA_PREFIX to the pixi env. + - 'export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"' + cluster_switch: train_path_c4: "/net/storage/pr3/plgrid/plggllmeffi3/datasets/c4/train" eval_path_c4: "/net/storage/pr3/plgrid/plggllmeffi3/datasets/c4/validation" train_path_fineweb: "/net/storage/pr3/plgrid/plggllmeffi3/datasets/fineweb/train" - eval_path_fineweb: "/net/storage/pr3/plgrid/plggllmeffi3/datasets/fineweb/train" + eval_path_fineweb: "/net/storage/pr3/plgrid/plggllmeffi3/datasets/fineweb/train" \ No newline at end of file